Xu Han

Recurrence's rapid learning of retrieval functions actively prevents attention from acquiring similar capabilities in hybrid architectures, necessitating staged training for optimal performance.

Aggressive KV cache compression can be achieved with near-zero accuracy loss by treating the cache as a trainable neural network optimized at test-time.

Deterministic forecasting embedded within generative models is unstable; DV-Flow decouples it as a parallel velocity, unlocking stable end-to-end training and superior probabilistic forecasts.